The cops and robbers problem:

The cops and robbers problem is a variation of a pursuit-evasion problem.
Cops and robbers are entities that exist on a graph of nodes. 
Taking turns, the cops and robbers may move to adjacent nodes. 
The goal of the game for the cops is to corner the robber such that any move
he makes would land him on a square shared by a cop. 
A robber wins if he can avoid capture indefinitely. 
In this scenario the cops would be moving non-deterministically,
so the robber should have a plan for evasion no matter what move the cops make.

Friendly robbers

An extra variation is the friendly robber
This robber wants to get captured which reverses the problem
Instead of trying to avoid all possible capture paths
You are trying to capture a randomly moving entity
The problem solution also reverses
The game ends if the robber is captured. To incentivise speed, the game randomly assigns the robber survived
If the robber survives the robber loses
Instead of most nodes leading to game terminated and a -1 meaning the robber loses
Most games end in -1 and a game terminated means the cop loses
No solutions can exist as all nodes have a chance to fail
Line_opposite_friendly.png shows a solution of the robber trying to get caught

benchmarks/single/Domain_generic.pddl

The domain file is commented for clarity. It is a problem-agnostic version of the domain with bounded size.
It is also much slower

Domains

problem_maker.ipynb can be used to create new domains programmatically
domains hard-code cop move actions to reduce state space

Problems

problem_maker.ipynb can be used to create new problems programmatically
Each problem has a different graph configuration and starting locations
Problems with 2 cops can take a long time to run, I recommend one robber unless the graph is simple (like a straight line)
If the robber has a solution a strong cyclic solution will be found
For example, the robber can avoid capture indefinitely on a square
If the robber is eventually caught, no strong cyclic solution will be found
For example, the robber loses on a straight line
Problem_line_chase.pddl is commented for clarity

graph_libraries

contains the data for generating graphs. the text files have adjacency graphs and the other files have 
the compressed data that can be extracted with showg
more graphs can be downloaded at http://users.cecs.anu.edu.au/~bdm/data/graphs.html

HOW TO RUN

Place project in directory of planner

launch docker
    docker run -it -v ${pwd}:/PROJECT rbp

example of generating benchmark
    python rbp/prp-scripts/evaluate.py --catalogue /PROJECT/cops_and_robbers/benchmark --collection collection_test --output /PROJECT/cops_and_robbers/benchmark/RESULTS --cache CACHE --planner rbp 

go to the benchmark and run
    cd cache 
    time python run.py

example of running individual instances
    rbp cops_and_robbers/benchmarks/single/d_decagon.pddl cops_and_robbers/benchmarks/single/p_decagon.pddl 
    rbp cops_and_robbers/benchmarks/graph7/d_graph7_441.pddl cops_and_robbers/benchmarks/graph_7/p_graph7_441.pddl --output-format 3

visualize the output
    prpviz policy.out

Planner not provided Not all domain and problem files pushed due to size